Gastrointestinal Malignancy in Iron Deficiency Anemia
Gastrointestinal Malignancy in Iron Deficiency Anemia
The aim of the study was to determine whether simple and objective clinical parameters could usefully identify subgroups of patients with IDA at particularly high and low risk of underlying malignancy of the GI tract. Given an overall prevalence of GI malignancy of about 10% in this patient population, it was decided—on a purely arbitrary basis—that identifying a low-risk subgroup with a cancer risk of <2%, and/or a high-risk subgroup with a cancer risk of >20%, might prove useful in clinical practice.
The study involved a retrospective analysis of clinical data for subjects on the Poole IDA clinic register assessed between June 2004 and May 2012 inclusive, who went on to have GI investigation and for whom information on the final diagnosis was available. Anonymised data were analysed using Predictive Analytics Software (PASW) to assess whether five clinical parameters—sex, age, haemoglobin concentration (Hb), mean red cell volume (MCV) and iron studies—could usefully predict the likelihood of GI pathology, and in particular malignancy, on subsequent investigation. Having viewed the protocol, the National Research Ethics Service advised that formal Research Ethics Committee approval was not required.
Age was categorised as a dichotomous variable (70 or less vs over 70 years), as the incidence of the major GI malignancies rises steeply after this age. An initial examination of the data revealed a striking stepwise increment in the prevalence of underlying cancer at around the age of 70 years—rising from 2.6% in the 60–69-years age group, to 11.6% in the 70–79-years age group. Hb and MCV were analysed as study population quartiles in order to obtain approximately equally sized groups, since there is no accepted way of categorising these variables—the quartiles are presented in order of increasing anaemia and microcytosis, respectively. Dichotomous outcome variables were created to indicate a diagnosis of malignant neoplasm, benign neoplasm over 1 cm in diameter (smaller lesions were felt unlikely to have made a significant contribution to the development of IDA), inflammatory disorder, vascular malformation or coeliac disease. A further outcome variable was created based on whether subjects had any of the above diagnoses.
Datasets were complete for sex, age, Hb, MCV and final diagnosis. Results for iron studies—transferrin saturation or serum ferritin or both—were available for 90% of the study population. The major reason for the absence of data for the remaining 10% was that iron supplements had been commenced prior to referral, without the relevant test being requested. Iron deficiency was therefore analysed as a dichotomous variable, being 'severe' (arbitrarily defined as a transferrin saturation <10%, a serum ferritin <10 μg/L or both) or 'non-severe' (criteria for severe deficiency not met).
Frequencies were run to provide descriptive information for the study population. Logistic regressions were used to investigate the associations with each explanatory variable on their own to obtain ORs, and together to adjust for all the other demographic and clinical characteristics. The parsimonious model was then selected for malignant neoplasm including only the factors that were significant (using a significance level of 5%) and 95% CIs were presented. The Nagelkerke R Square was used to assess the proportion of variance explained in the data, that is, a measure of how well the models fitted the data.
Method
The aim of the study was to determine whether simple and objective clinical parameters could usefully identify subgroups of patients with IDA at particularly high and low risk of underlying malignancy of the GI tract. Given an overall prevalence of GI malignancy of about 10% in this patient population, it was decided—on a purely arbitrary basis—that identifying a low-risk subgroup with a cancer risk of <2%, and/or a high-risk subgroup with a cancer risk of >20%, might prove useful in clinical practice.
The study involved a retrospective analysis of clinical data for subjects on the Poole IDA clinic register assessed between June 2004 and May 2012 inclusive, who went on to have GI investigation and for whom information on the final diagnosis was available. Anonymised data were analysed using Predictive Analytics Software (PASW) to assess whether five clinical parameters—sex, age, haemoglobin concentration (Hb), mean red cell volume (MCV) and iron studies—could usefully predict the likelihood of GI pathology, and in particular malignancy, on subsequent investigation. Having viewed the protocol, the National Research Ethics Service advised that formal Research Ethics Committee approval was not required.
Age was categorised as a dichotomous variable (70 or less vs over 70 years), as the incidence of the major GI malignancies rises steeply after this age. An initial examination of the data revealed a striking stepwise increment in the prevalence of underlying cancer at around the age of 70 years—rising from 2.6% in the 60–69-years age group, to 11.6% in the 70–79-years age group. Hb and MCV were analysed as study population quartiles in order to obtain approximately equally sized groups, since there is no accepted way of categorising these variables—the quartiles are presented in order of increasing anaemia and microcytosis, respectively. Dichotomous outcome variables were created to indicate a diagnosis of malignant neoplasm, benign neoplasm over 1 cm in diameter (smaller lesions were felt unlikely to have made a significant contribution to the development of IDA), inflammatory disorder, vascular malformation or coeliac disease. A further outcome variable was created based on whether subjects had any of the above diagnoses.
Datasets were complete for sex, age, Hb, MCV and final diagnosis. Results for iron studies—transferrin saturation or serum ferritin or both—were available for 90% of the study population. The major reason for the absence of data for the remaining 10% was that iron supplements had been commenced prior to referral, without the relevant test being requested. Iron deficiency was therefore analysed as a dichotomous variable, being 'severe' (arbitrarily defined as a transferrin saturation <10%, a serum ferritin <10 μg/L or both) or 'non-severe' (criteria for severe deficiency not met).
Frequencies were run to provide descriptive information for the study population. Logistic regressions were used to investigate the associations with each explanatory variable on their own to obtain ORs, and together to adjust for all the other demographic and clinical characteristics. The parsimonious model was then selected for malignant neoplasm including only the factors that were significant (using a significance level of 5%) and 95% CIs were presented. The Nagelkerke R Square was used to assess the proportion of variance explained in the data, that is, a measure of how well the models fitted the data.